MNEMONIC ADVANTAGESOF THE SELF
Effects of Manipulation Class and the Nature of
Processing Manipulation class
. The first moderator that affected the
magnitude of the SRE across the literature was that a larger mean SRE was obtained in SR-semantic studies than in SR- OR studies. First, note that the SRE occurs on the average within both manipulation classes, although study findings were inconsistent. This finding begins to suggest possible differences between person-reference versus semantic processing as tasks, a topic that we discuss below.
Within the class of studies that used SR-OR manipulations, given the debates that have occurred in the literature, the finding that SR tends to promote better memory than OR may be surpris- ing (cf. Bower & Gilligan, 1979; Brown et al., 1986; Lord, 1980, 1987; Maki & McCaul, 1985). Those debates arose be- cause of inconsistent study findings; for example, reference to familiar others apparently causes the SRE to "disappear," lead- ing researchers to argue that SR creates no memory advantage over reference to a familiar other. The results of our study show that, although the SRE was weaker in SR-OR studies, it was nevertheless present and statistically significant, as predicted.
The finding that the SRE was larger for SR-semantic manipu- lations is also intriguing if one begins to consider the nature of the underlyingmemory processes and structures. If, for example, it is assumed that both self and semantic representations (or processing, for that matter) are qualitatively similar, then one must also be able to explain why SR tends to facilitate memory better than semantic encoding overall. None of the models that we analyzed could completely explain the variation in the SR- semantic cases. Although we must be careful about conclusions with regard to the structures that underlie SR and semantic processing that are based on correlational evidence, one plausi- ble conclusion is that SR is more effective in producing good recall than is a semantic task for a number of reasons. Specifi- cally, the typical SR task (a) uses traits as stimulus items more than 80% of the time, (b) taps trait domains likely to have been elaborated on many times using SR, (c) is likely to promote an SR mode in retrieval because of this practiced elaboration of traits, and (d) taps trait domains that are, because of the two points just mentioned (b and c), likely to be highly organized along self-related (or at least person-referent) dimensions. Con- sistent with this reasoning, SR promotes better recall than se- mantic processing across the literature; but, perhaps more im- portant, when we examined only SRE studies that used SR and semantic tasks that promote both relational and item-specific processing, the SRE was much smaller than it was for the set of all SR-semantic cases, k = 7, d = 0.29, 95% CI = 0.09- 0.49. However, even in this subclass of studies, there was sig- nificant heterogeneity, Qwi(6) = 22.41, p < .01; thus, although our codings of relational and item-specific processing predict some of the variation in SR-semantic studies, it does not ac- count for all of the variation (cf. Klein & Loftus, 1988). As we discuss later, some models we examined within the SR- semantic class help to shed light on important moderators of the relationship, however, because in almost every case study findings were inconsistent, primary-level research is still needed to isolate the causal mechanisms responsible for the difference between SR and semantic encoding.
Model test for nature of processing induced. The second important model test shows that the degree to which compared encoding tasks are equivalent in promoting item-specific and relational processing affects the magnitude of the SRE. As stated earlier, for each study in the literature, the comparison tasks were judged to promote either relational processing, item-spe- cific processing, or both (e.g., "Does the word mean the same as xxx?" probably promotes elaborative processing predominantly, whereas "Does the word describe your mother?" likely pro- motes both kinds of processes). Results show significantly smaller SREs on average for tasks that involved both kinds of processing. This would be expected if SR and the comparison task both promote the same kinds of processing. In addition, the effect sizes for this class were consistent, thus no further model testing is needed to explain that variation. Nonetheless, it is important to note that, even when the comparison task involved both kinds of processes, there was still a significant SRE for that class of studies. Presumably, SR not only invokes both processes but possibly more relational and item-specific processing. Of course, other alternatives are possible: For exam- ple, some additional process or condition may be invoked that results in superior memory, such as encoding specificity, or, because the majority of studies in this class used trait adjectives, it may be that SR poses an advantage because of the sheer frequency with which the average person relates traits to himself or herself (Wells et al., 1984).
Also interesting, and not wholly unexpected, is the finding that there was no difference in the magnitude of the SRE for studies in which the comparison task involved either relational or item- specific processing. In both cases, SR was superior in producing memory because theoretically SR invokes both. In both classes, however, there was significant variation in effect sizes. 'However, based on our data, the process of controlling for organization or elaboration does not entirely eliminate the SRE.
Roles of Organization and Elaboration
Understanding the role of organization can help us to explain the difference between semantic processing and SR if we take some of the following points into account. We have identified studies in which the comparison task has the potential to invoke relational (vs. item-specific) processing; however, just because a task promotes relational processing does not mean that mem- ory will necessarily be facilitated. For example, the provision of cues to help recall a category may facilitate memory for the category but not necessarily items in the category (Tulving & Pearlstone, 1966). In addition to the possibility that organiza- tion will not facilitate memory in all circumstances, we note too that, except for a few instances in the SRE literature, organi- zational instructions to the participant are rarely used in the design of studies (for notable exceptions, see Klein & Kihl- strom, 1986; and Klein & Loftus, 1988). Nonetheless, Klein and his colleagues have produced evidence that some SR tasks naturally promote such organization and that such organization does facilitate memory. Other evidence suggests that people will use categorization in the absence of category labels (e.g., Lewis, 1971), and that category size is an important task parameter. For example, Hunt and Seta (1984) showed that large categories were more likely to promote relational processing. However,